基于融合特征的自适应阈值镜头边界检测算法

Adaptive threshold shot boundary detection algorithm based on fusion features

针对目前镜头边界检测算法易造成错检漏检、人工确定阈值具有不确定性及渐变镜头相邻帧之间特征变化较小难以检测到的情况,提出一种融合RGB颜色直方图特征与方向梯度直方图(HOG)特征的自适应阈值多步比较方案镜头边界检测算法。通过计算多个步骤的帧之间的差异,生成一个多步帧差模式距离图,分析它们在模式距离图中的模式来检测它们的变化,在阈值确定方面加入自适应阈值。采用卷积神经网络(CNN)对视频帧提取特征,使用该算法与单一特征算法及其它文献算法作比较。实验结果表明,该算法的查全率和查准率相比其它算法都有较好提高。

In view of the situation that the current shot boundary detection algorithm is easy to cause error detection,leakage detection and difficulty to detect the feature change between adjacent frames of gradient shot,a shot boundary detection algorithm based on RGB color histogram feature and histogram of oriented gradients(HOG)feature was proposed.By calculating the differences between frames of multiple steps,a frame difference mode distance graph was generated,and their patterns in the pattern distance diagram were analyzed to see if they were gradients or abrupt shots.An adaptive threshold was added to determine the threshold.Convolution neural network(CNN)was also used to extract the features of video frames.The proposed algorithm was compared with single feature algorithm and other literature algorithms.Experimental results show that the recall rate and precision rate of this algorithm are better than other algorithms.